=Paper=
{{Paper
|id=Vol-2524/paper15
|storemode=property
|title=A virtual counselor for online social networks (or did I really want to send you my post?)
|pdfUrl=https://ceur-ws.org/Vol-2524/paper15.pdf
|volume=Vol-2524
|authors=Francesco Orciuoli,Mimmo Parente
|dblpUrl=https://dblp.org/rec/conf/psychobit/OrciuoliP19
}}
==A virtual counselor for online social networks (or did I really want to send you my post?)==
https://ceur-ws.org/Vol-2524/paper15.pdf
A Virtual Counselor for Online Social Networks
(or did I really want to send you my post?) ?
Francesco Orciuoli1 and Mimmo Parente1
Dip.to Scienze Aziendali - Management and Innovation Systems
Università degli Studi di Salerno
Italy
{forciuoli,parente}@unisa.it
Abstract. It is well established by the scientific literature that a great
part of Face-to-Face communication occurs at a non verbal level and
also that this is partly obfuscated when the interaction takes place by
Computer-Mediated-Communication (CMC). Fully aware we are simpli-
fying the subject, we can say that the communication satisfies the need
of humans to share their emotional experiences. Thus we can say that
also the formulation and interpretation of messages exchanged in CMC
is influenced by the emotions. Moreover the reduced physical presence
in CMC implies a lack of social norms or social control which is ampli-
fied in the Online Social Networks (OSN) scenario. Motivated by these
naı̈ve considerations and the massive use of OSN, we introduce a Virtual
Counselor as a contribution, from a technical point of view, to augment
the quality of communication among users of OSN. The implementation
of this Virtual Counselor is based on technologies by now mature like
wearable devices to measure physical parameters, on artificial emotional
intelligence, and on interactive tutoring systems, strongly used in online
learning environments.
We propose an abstract model of the communication scenario in OSN
containing the Virtual Counselor to help the interpretation of the mes-
sages and of the emotional states in order to improve the communication
among parties. The goal is to align the emotional states of senders and
receivers to form dynamic groups of target friends in the OSN to send the
posts to. The dynamism of the groups is both spatial (that is the com-
position of receivers can change for a given sender and a given message)
and temporal (for example the sending of a message can be postponed
in time).
We think that this model is the basis for defining a new class of tools to
improve the communication in OSN.
Keywords: Computer Mediated Communication · Online Social Net-
works · Emotion Representation · Artificial Emotional Intelligence.
?
Copyright c 2019 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
2 F. Orciuoli and M. Parente
1 Preliminaries
Positive and negative effects of massive use of technology have been thoroughly
investigated by the scientific community (see e.g. [14]), but natural worries af-
fects tech-scientists, and common sense people as well. In particular the over-
whelming use of Online Social Networks (OSN) has changed the perception
of the communication among individuals. All of us experience every day how
Computer-Mediated Communication (CMC) through OSN sometimes is more
difficult compared to the Face-To-Face communication. In this latter setting the
nonverbal cues are a great part of the communication, see e.g. [6], though in
CMC it has been shown that the emotions are not absent and are not even
difficult to communicate, see [2].
In this paper we assert that the communication as is today in OSN occurs too
often without being fully aware of the role, or better of the consequences, that
the emotions can play. In particular in CMC usually there is no awareness of
what are the post-effects of both the emotional state of the sender that in some
way leaks through the messages we sent and of the emotional state of the receiver
when she/he interprets the received message. The way the communication occurs
today on OSN is, in a sense, blind. Meaning that a user u posts to all her
friends/followers without having knowledge of the emotional state of her reader,
but also without exposing her state. A consequence of this is that the receiver
might misunderstand the original meaning of u’s post.
Here we propose a model for a communication scenario in which there is a
Virtual Counselor which can interpret the emotional state both of the senders
and the receivers and of the message content as well. From this interpretation,
the Virtual Counselor can give explicit suggestions both to receivers and senders
to improve the interpretation of the message or even to determine which is the
right subset of friends/followers to send the message or also to suggest when and
if to send it.
The paper is organized as follows: in section 2 we sketch the model of the
Virtual Counselor, in section 3 we discuss about technologies issues to implement
the model and in section 4 we give some conclusions.
2 The model
The approach we propose starts from the model of Garrod and Pickering [9]
represented in Fig. 1. Such a model clearly represents that in order to align the
meaning of a message or, in general, of a dialogue it is needed to employ the
work of at least four levels (from the bottom):
– the physical layer
– the syntactic-lexical layer
– the semantic layer
– the model layer
A Virtual Counselor for Online Social Networks 3
Fig. 1: Model of Garrod and Pickering [9]
The emotions may hinder the alignment task provoking negative side effects.
The proposed approach focuses on improving the quality of communications in
OSN and is based on the idea that it is possible to inject in such environment
a Virtual Counselor, shortly VC. This latter resembles the Intelligent Tutoring
Systems (ITS), often applied in e-learning scenarios, [12]. In particular, tradi-
tional ITS are software systems providing adaptive educational experiences for
both students and teachers. Adaptivity is supported by ITS in several ways: i)
providing students’ learning activities coherently with their current knowledge
and skills, in order to foster meaningful learning; ii) providing individualized
feedbacks, able to stimulate next learning activities and avoid frustration, de-
motivation and disengagement due to unsuccessful performances; iii) providing
hints helping students during the execution of their learning tasks. Therefore,
our aim is to introduce a sort of adaptive tutor, dealing with emotions, in a com-
munication context (in the place of the learning context), where the task is not
learning but communicating, and in which we have not learners but participants
to a message exchanging activity. Its main contributions are: i) supporting the
sender in selecting a suitable group of receivers with respect to the emotional
aspects, ii) providing alerts to the sender when his/her emotional state and the
emotions expressed in his/her text message are not compliant with the context of
the discussion, iii) prompting the receiver to support the right comprehension of
the message, and iv) suggest the sender to possibly postpone in time the sending
of the message. In our model, see Fig. 2, the VC constructs its knowledge about
the emotional states of senders and receivers, extracts affective elements from
the messages (outgoing or incoming) and applies rules to suggest emotion-based
4 F. Orciuoli and M. Parente
Fig. 2: Sketch of the proposed approach.
actions to reduce misunderstanding and negative side effects in communication
tasks in the OSN.
In order to implement our model, we need to represent emotions as discrete
elements belonging to a set E. Therefore, it is useful to refer to the Plutchik’s
discrete emotion model, [11], that maps eight basic emotions into a wheel: joy,
trust, fear, surprise, sadness, disgust, anger and anticipation, see Fig. 3, for a
simplified version. An alternative might be represented by the model proposed
by Lang [7] that categorizes emotions in a 2D space by valence and arousal, see
Fig. 4. For example, anger has negative valence and high arousal while sadness
has negative valence and low arousal.
In order to define the behaviour of the VC we now define the communication
scenario in which it operates.
Definition 1. Given an alphabet Σ, a set of users U and a set of emotions E,
a Communication Scenario C is a 6-tuple C = (S, R, t, ctx, σh , σm ) where1
– S ⊆ U is the set of senders;
– R ⊆ U is the set of receivers;
– t ∈ Σ ∗ is a text-message string;
– ctx is the context in which the communication takes life (e.g., the thread of
the discussion, the topics of the dialogue);
1
Given a set A, the symbol P(A) denotes the power set of A, that is the set of all
subsets of A.
A Virtual Counselor for Online Social Networks 5
– σh : U → P(E) returns the emotions of a user u ∈ U ;
– σm : Σ ∗ → P(E) returns the emotions expressed in t.
In what follows, for simplicity, we will consider only the case where |S| = 1, that
is in the communication scenario there is only one sender for each message. The
functions σh and σm return the emotions of the humans (the users) and of the
messages, respectively.
Actually, the values of the functions σh are weighted with a value between 0
and 1 which expresses the confidence of the computed value. If this value is too
low, a confirmation can be asked to the sender or to the receiver.
For the moment we will just give an hint on how the functions can be com-
puted in section 3, in the extended version of the paper we will formalize the
scenario.
Given the values returned by the emotional functions, the composition of
these values determines the support for the decision or, better the suggestion of
the Virtual Counselor.
Fig. 4: 2D emotion space model of
Fig. 3: The Plutchik wheel Lang
Definition 2. A suggestion g is a statement of the following form:
g := {recommended action} · {object} · {explanation}
For the moment let us give some examples of statements.
Definition 3. Given a Communication Scenario C = (S, R, t, ctx, σh , σm ) and
a set of emotions E, a Virtual Counselor for C is a mapping
VC u : P(E)|S|+|R|+1 → G
that computes the suggestion g ∈ G for the user u, based on the emotions of the
|S| senders, |R| receivers and the emotion extracted from t.
6 F. Orciuoli and M. Parente
Recommended action Object Explanation
send t to R̄ since emotional state of the sender/receivers
are compliant
reflect on t and ctx since emotional state of the sender and the
(before sending) emotions in the message are not compliant
with the context of the communication
evaluate better t and ctx since the emotional state of the receiver is not
compliant with the emotions expressed by the
sender
confirm or deny t since emotional state of the sender and the
(before sending) emotions in the message are not compliant
In words, the Virtual Councelor for a user u, is a function with |S| + |R| + 1
arguments: the first |S| arguments represent the emotions of the senders, the
other |R| represent the emotions of the receivers and the last is the emotions
of the message. Let us underline that the user u can play both the role of a
sender or of a receiver, thus the VC behaviour is polymorphic and the role of u
is indicated by u↓ , when u plays the role a receiver, and u↑ when a sender.
Let us illustrate the framework above with an example.
Example 1. Let U = {s, r1 , r2 , r3 }, where s is the sender of the message t and
ri , i = 1 . . . 3, are the receivers (friends/followers in a OSN). The emotional
functions are defined as follows: σh (s) = {excited}, σh (r1 ) = {pleased}, σh (r2 ) =
{sad}, σh (r3 ) = {glad} and σm (t) = {amused}.
In this case, it is possible for the Virtual Counselor to suggest the exclusion
of r2 from the list of the receivers of the message t, given that the emotional
state of s and the emotions expressed by t are not compliant with the emotional
state of r2 . In this scenario, the compliance of the emotions is obtained by using
the 2D emotion space model of Lang [7] and reported in Fig. 4. The suggestion
is then:
Recommended action Object Explanation
send t to {r1 , r3 } since emotional state of {r1 , r3 }
are compliant with the one of s
while that of {r2 } is not.
3 Technological issues
In order to implement the functions defined in the previous section, we need to
employ three technological assets.
In particular, σm can be supported by the first technological asset that is
sentiment analysis. Sentiment analysis [1] is a very interesting field of Natural
Language Processing. Sentiment analysis can be used to detect emotions in text
messages exchanged between sender and receivers. Many existing approaches fo-
cused on word-level analysis of texts and are able to detect only explicit expres-
sions of sentiment. Other techniques, recognized in literature, have been defined
A Virtual Counselor for Online Social Networks 7
to deal with emotions that are not expressed in the text by using words with an
affective meaning. In these cases, the text describes real-life situations, which the
reader is able to associate to specific emotions by using his/her commonsense
knowledge.
Moreover, σh can be supported by the second technological asset that is based
on the use of the physiological signals [13], which include the electroencephalo-
gram (EEG), temperature (T), electrocardiogram (ECG), photoplethysmogra-
phy (PPG), electromyogram (EMG), galvanic skin response (GSR), respiration
(RSP), hearth rate (HR), hearth rate variability (HRV), to recognize emotions.
In particular, studies like [8] assert that heart-related parameters, like HRV, can
be exploited to detect emotions through the unobtrusive help of smart wearable
devices and the application of machine learning techniques, typically for classi-
fication tasks. Moreover, HRV can be extracted from PPG [10] that is widely
used in smart watches for this aim. A PPG sensor is non-invasive, in fact it uses
a light emitting diode and a photo-diode to record the pulse waveform. More
in details, the authors of [4] selected 5 features (Coefficient of variance of the
pulse peak intervals, Respiratory sinus arrhythmia, Oscillation of the baroreflex,
Ratio of low and high frequency, Standard deviation of the pulse intervals) from
a total of 13 features to recognize 5 emotions (sad, angry, fear, happy, relax).
The aforementioned paper reports that the highest accuracy has been achieved
using a SVM classifier.
The third asset supports both the functions and consists in a computa-
tional ontology for semantically describing affective phenomena such as emo-
tions, moods, appraisals and subjective feelings [3] [5]. In our approach, the
ontology for emotions can be used to support classification tasks and to reason
on emotions detected from both text messages and emotional states of sender
and receivers.
4 Conclusions
We have presented, in a very preliminary way, our idea to improve the commu-
nication among users in OSN. The idea stems from the pervasive and massive
and daily use of OSN to communicate. Often the communication leads to mis-
understanding among the users due to a hasty writing of the message sometimes
in a particular emotional state. Based on these considerations, we think that the
users could be supported by a Virtual Counselor that can be implemented by
means of mature technologies like sentiment analysis, NLP techniques, machine
learning, and wearable devices.
A step forward of our research is to consider also feedback of the user’s
posts and apply learning techniques to improve the Virtual Counselor suggestion
quality.
References
1. Balahur, A., Hermida, J.M., Montoyo, A.: Detecting implicit expressions of
emotion in text: A comparative analysis. Decision Support Systems 53(4),
8 F. Orciuoli and M. Parente
742 – 753 (2012). https://doi.org/https://doi.org/10.1016/j.dss.2012.05.024,
http://www.sciencedirect.com/science/article/pii/S0167923612001352, 1) Compu-
tational Approaches to Subjectivity and Sentiment Analysis 2) Service Science in
Information Systems Research : Special Issue on PACIS 2010
2. Derks, D., Fischer, A.H., Bos, A.E.: The role of emotion in computer-
mediated communication: A review. Computers in Human Behavior 24(3),
766 – 785 (2008). https://doi.org/https://doi.org/10.1016/j.chb.2007.04.004,
http://www.sciencedirect.com/science/article/pii/S0747563207000866, instruc-
tional Support for Enhancing Students’ Information Problem Solving Ability
3. Gil, R., Virgili-Gomá, J., Garcı́a, R., Mason, C.: Emotions ontology for collabora-
tive modelling and learning of emotional responses. Computers in Human Behavior
51, 610 – 617 (2015). https://doi.org/https://doi.org/10.1016/j.chb.2014.11.100,
http://www.sciencedirect.com/science/article/pii/S0747563215001417, computing
for Human Learning, Behaviour and Collaboration in the Social and Mobile Net-
works Era
4. Guo, H.W., Huang, Y.S., Lin, C.H., Chien, J.C., Haraikawa, K., Shieh, J.S.: Heart
rate variability signal features for emotion recognition by using principal compo-
nent analysis and support vectors machine. In: 2016 IEEE 16th International Con-
ference on Bioinformatics and Bioengineering (BIBE). pp. 274–277. IEEE (2016)
5. Hastings, J., Ceusters, W., Smith, B., Mulligan, K.: The emotion ontology: En-
abling interdisciplinary research in the affective sciences. In: Proceedings of the
7th International and Interdisciplinary Conference on Modeling and Using Con-
text. pp. 119–123. CONTEXT’11, Springer-Verlag, Berlin, Heidelberg (2011),
http://dl.acm.org/citation.cfm?id=2045502.2045516
6. Knapp, M.L., Hall, J.A., Horgan, T.G.: Nonverbal communication in human inter-
action. Cengage Learning (2013)
7. Lang, P.J.: The emotion probe: studies of motivation and attention. American
psychologist 50(5), 372 (1995)
8. Maria, E., Matthias, L., Sten, H.: Emotion recognition from physiological signal
analysis: A review. Electronic Notes in Theoretical Computer Science 343, 35–55
(2019)
9. Pickering, M.J., Garrod, S.: Toward a mechanistic psychology of
dialogue. Behavioral and Brain Sciences 27(2), 169–190 (2004).
https://doi.org/10.1017/S0140525X04000056
10. Pinheiro, N., Couceiro, R., Henriques, J., Muehlsteff, J., Quintal, I., Goncalves, L.,
Carvalho, P.: Can ppg be used for hrv analysis? In: 2016 38th Annual International
Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). pp.
2945–2949. IEEE (2016)
11. Plutchik, R.: The nature of emotions: Human emotions have deep evo-
lutionary roots, a fact that may explain their complexity and pro-
vide tools for clinical practice. American Scientist 89(4), 344–350 (2001),
http://www.jstor.org/stable/27857503
12. Polson, M.C., Richardson, J.J.: Foundations of intelligent tutoring systems. Psy-
chology Press (2013)
13. Shu, L., Xie, J., Yang, M., Li, Z., Li, Z., Liao, D., Xu, X., Yang, X.: A review of
emotion recognition using physiological signals. Sensors 18(7), 2074 (2018)
14. Turkle, S.: Alone together: Why we expect more from technology and less from
each other. Hachette UK (2017)